Himanshu Kalra

Deeply personalised learning

Sat, 04 Jul 2026 00:00:00 +0000

One thing AI has unlocked for me is deeply personalised learning. Multiple books/papers synthesized into one linear learning sequence - broken down into atomic notes, Q&A flashcards, quizzes - uploaded to netlify in a format readable across devices auto-saving my progress.
If I am reading business advice, it is tailored to my unique goals and situation. If I am learning with someone else, there is a scoreboard. Spaced repetition and active recall built into the format.
-
Used Fable as the manager+architect with GPT 5.5 (via Codex CLI) writing and Opus 4.8 reviewing - all coordinated through Claude CLI

Learn and teach

Wed, 20 May 2026 00:00:00 +0000

The core of a User Researcher's job boils down to two things: Learn and Teach.
Learn about users. Teach what you've learned to the builders and decision-makers.
What makes a researcher effective at this? Here's a take:
1/ Teach only some of what you learn - and do it strategically. No infodumping. Give people only what they need for the next decision. Plant seeds early for decisions that are still downstream.
2/ Your students outrank you, and they didn't sign up for class. This is unlike any other teaching relationship. You're correcting and informing people with more power and organizational visibility than you. The authority you carry is only what they've chosen to extend - which means you're always earning it, never holding it.
3/ Expect disengagement - and learn to make anything interesting. The question is always: what does this particular person care about? How does what you found connect to the problem they're facing? If there's no connection, they probably don't need to know.
4/ Repetition is how ideas move through organizations. Not by saying the same thing twice - but by returning to the same core insights from different angles, in different rooms, with different audiences over time.
All of this requires genuine caring - for the people using the product, and for the people building it.
Which makes the current moment genuinely hard. If you're worried your role will be automated away or treated as optional, if you don't feel safe pushing back - it's difficult to show up as a passionate teacher. It's hard to care well for others when your own footing is uncertain.

It's just an AI wrapper

Fri, 15 May 2026 00:00:00 +0000

"It's just an AI wrapper." Said like an insult
Nobody looks at a drill and says "it's just a motor." Nobody calls a fan "just a motor with blades." But that's exactly what those things are.
It's just that someone put the effort to figure out the grip angle. The speed settings. The cord length for a real kitchen. The noise level for a real bedroom. The switch placement for someone doing it one-handed.
That work - the unglamorous, obvious-in-hindsight work of fitting a capability to a real human need - is what turns a technology into a tool.
AI is the motor. Powerful, general, indifferent.
The wrapper is what makes it useful to the person who just needs to get something done and doesn't care how the thing works.
So yes. Wrap the model. But put the effort to understand your people.
The moat lives in that intimacy.

AI changes the question

Wed, 13 May 2026 00:00:00 +0000

Most security product research answers one question: can analysts find and triage an alert, run a query, or investigate an endpoint? Most mature products have solved this.
AI changes the question. When an analyst dismisses an alert because the AI flagged it low-priority, did they evaluate it independently or accept the recommendation? Defer too readily and you miss what the model wasn't trained on. Ignore the recommendations and you've bought automation you don't use.
Both look the same from the outside - an analyst made a call and moved on. Neither shows up until something is missed.
Research on human-AI collaboration has established methods for studying this - trust calibration, appropriate reliance, confidence elicitation - and applied them in clinical contexts. They haven't been applied to SOC workflows. AI triage platforms are scaling now. Radiology went through this - AI-assisted image reading raised the same question: when does the radiologist trust the flag, when do they override it? That's now an active research area with published methods.
If security vendors are doing this work, they're not publishing it. Which means it's a competitive edge - or a gap. Probably both.

A reckoning with complexity

Wed, 22 Apr 2026 00:00:00 +0000

One hopeful consequence of AI: it may force a reckoning with complexity. When you can generate a convincing argument for almost any position - and against it - the question "what is the right thing to do?" loses its clean answer.
Do you take the flight to close the deal, knowing your company has a net-zero pledge? The carbon is real. So is the relationship that won't form over a Zoom call.
Do you recommend the automation that cuts costs - and three jobs? Efficiency serves the many. Displacement lands on the few.
Do you tell a colleague their leadership style is holding the team back? Honesty serves them. It might also just serve you.
That discomfort - not knowing which side is right - might be the opening we need.
A model of Human Development - Spiral Dynamics - calls this Stage Yellow: the ability to hold multiple valid frameworks at once and still make a call. Not that all answers are equal, but that all frameworks/models are contextual and should be held loosely, functionally.

Facilitating insights, not delivering them

Thu, 20 Nov 2025 00:00:00 +0000

When generating insights becomes faster, the next frontier might be *facilitating* them
The thing is: Researchers aren't the decision makers. We learn from users and carry that information to the people who make decisions. But traditional delivery often fails. Research repos collect dust. New frameworks promise to make insights "stick" but rarely do
The reason is simple: transfer of information isn't transfer of conviction. What's learned without being felt is forgotten. People are most convinced by reasons they discover themselves. (Pascal)
So maybe the standard practice will move from *finding and delivering* to *finding and facilitating derivation* - like teachers helping derive a math equation - the logical chain is established and the conclusions are clear, but re-deriving together helps the learner foster a stronger sense of ownership and understanding compared to sharing beautifully-packaged final equations (or insights)
(We've seen this work in workshops. The question is how we make facilitation the default, not a once-in-a-while technique reserved for highly important and visible studies)

Steel-manning research findings

Thu, 30 Oct 2025 00:00:00 +0000

Notes on steel-manning user research findings:
1/ risk and impact. what are they risking if they don't act on it. what can they gain if they do. great to include numbers that measure the impact. estimates, quick arithmetic - and walking them through them - is far better than nothing
2/ stories. why it matters for the user. in what situation would it be crucial. drama - emotional highs and lows - is a part of human life, and should be a part of stories you tell
3/ prioritise it for them. PMs already have quarterly planning. Where might the findings map onto what's already planned?
4/ behavioral AND attitudinal data - made sense of and narrated to give insight into ongoing product/business issues and what may solve them

First principles of user research

Mon, 10 Mar 2025 00:00:00 +0000

When entering unfamiliar territory, it's useful to keep returning to the first principles of User Research:
1/ Start with the assumption that you don't know what users need or how they think. (Epistemic Humility)
2/ What people say often differs from what they do.
3/ Many usage behaviors rely on unspoken, embodied knowledge that users cannot articulate. Research must surface this tacit dimension.
4/ User behaviors are shaped by their environments, tools, social contexts, and constraints - not just by preferences or rational choices.
5/ Research findings are only meaningful to the extent that they represent users' actual contexts, environments, and behaviors. The lab is never the same as the wild.
6/ Users can reliably identify problems they experience but are less reliable at envisioning effective solutions. Their struggles are data; their suggestions are hypotheses.
7/ The models we create of users and their behaviors are always simplifications. The map is not the territory.
8/ Focus research on the zone just beyond what you already know, where uncertainty is highest and learning potential is greatest.
9/ Research can never be comprehensive. Focus on reducing uncertainty in decision-critical areas rather than attempting exhaustive understanding.
10/ Package insights differently for different audiences (designers, engineers, executives) without compromising integrity or nuance.

Learning flows both ways

Tue, 31 Dec 2024 00:00:00 +0000

The interesting part about AI interaction is that the usage style is unique to each person and learning flows both ways.
AI capabilities are improving quickly and users' mental models are shifting with it. What works in Week 1 might be irrelevant by Week 4.
In this evolving relationship, designers need to actively understand:
> how people ascribe agency, intelligence and intentionality to AI responses, and how that affects their interaction style and emotional investment
> how people develop working models of AI reliability and capabilities, similar to how we form attachment styles through repeated interactions and expectation-setting
> what makes people feel comfortable sharing personal information, and how this affects depth of engagement and learning outcomes
With AI agents being a significant focus area of big tech in 2025, perhaps we will shift our focus from designing a sequence of interactions to designing dynamically evolving relationships.

Andrea Knight Dolan on AI in research

Tue, 30 Jul 2024 00:00:00 +0000

An 18-year Google veteran,
Andrea Knight Dolan
is a leading expert in applying AI to UX research. She offers consulting, teaching, and mentorship services to help organizations harness the power of AI for better user experiences.
Here's what I learnt from her discussion with
Christine Perfetti
last week:
1/ AI enables scale - and raises expectations
"With AI moderators, it's not 10 interviews anymore. It's 30. It's not 30 interviews total. It's 30 per country. As the magnitude of data increases, it is where the opportunity and the necessity to rely on AI to process the data comes in."
2/ There are a range of products available (general and specialised), each with it's use case for Research. In a recent project, she used:
ChatGPT-4o for:
- Domain research and literature reviews
- Brainstorming and drafting interview scripts
- Translating responses (e.g., Spanish to English)
- Summarizing respondent answers
- Cross-checking responses across interviews
Voicepanel
for AI-moderated interviews in Spanish (a language she doesn't speak💡)
AddMaple
(Quant analysis tool) for Open-ended coding and thematic tagging. "AddMaple did in 3 minutes what would usually take me 45-50 mins of manual work. And those 3 minutes were reviewing it's work"
CoLoop
for segment analysis
Reveal
for hypothesis testing with PII stripping
3/ Synthetic (AI-generated) user data may be convincing, but they do NOT represent real user behaviour. Use it carefully.
AI has never truly lived - it hasn't slept, exercised, or eaten, and it can only regurgitate existing discussions, and not truly respond to new stimuli. Using an AI-generated user is like modeling an expert user who knows exactly what to do and don't make any mistakes.
4/ If you are a skeptic, a safe and effective way to start is to use AI for Desk Research
Use ChatGPT as your resource librarian, ask it to quickly find relevant resources and summarize key points. If you find any good reports, share it with GPT to chat with them
5/ Micromanage the tools until you trust them
Treat AI tools as research assistants and give them specific, targeted tasks instead of broad assignments. Maintain control and oversight, ensuring the quality and relevance of insights while leveraging AI's efficiency for tasks like data processing and analysis.
"You don't ask it to boil the ocean. You ask, 'This is what the stakeholder needs to know. Do we have evidence of this here?'"
6/ The most important skill for UX Researchers? Interviewing.
AI tools help scale up the extent and speed up the process, yet if you don't ask the right questions, you don't collect the right information. Even with an AI moderator, you have to program the interview, and you have to know if it's doing a good job. The classic principle still holds: Garbage in, garbage out.

When everyone can create

Sun, 05 May 2024 00:00:00 +0000

GenAI tools raise everyone's ability to create. Now, more and more people can create with less and less skills and effort.
You no longer *need* to know:
- color theory to create illustrations, or
- data structures and algorithms to write a program, or
- the difference between inductive and deductive reasoning to write a research plan
(Hurray for reducing the grunt between idea and output)
But when it's easier to create, it seems that the value of human discernment will go up.
Discernment for:
- Quality: Mediocrity is easy and aplenty, high quality work stands out. Domain expertise helps.
- Relevance: Is it relevant and useful for the people it is created? Context and cultural sensitivity helps. (Shameless plug for the importance of UX Research)
- Facts v/s Biases(Human) and Hallucinations (GenAI): Both Humans and AI are fallible. Critical thinking and evaluation helps.
#genai
#creativity

Books aren't sacred

Fri, 22 Sep 2023 00:00:00 +0000

Books aren't sacred.
They are objects like any other.
Ink on paper. Dead weight. Collecting dust.
A book is only valuable to the degree you engage with it.
Argue with the author.
Wrestle with the ideas.
Test them for yourself.
Keep what works, discard the rest.
Books aren't sacred.
Scribble on them with pens and markers.
Rip your favorite parts and paste them on your desk.
By themselves, books are nothing but dead wood.
Engaging, embodying and expressing their ideas is what keeps them alive.